Parametrization of Gaussian Approximation Potential for the Global Optimization of Magnesium Clusters MgN (N≤100)

Abstract

A two-stage GridSearch combined with active learning was employed to optimize GAP model parameters for Mg clusters, enabling reliable structural predictions in the extrapolative domain Mg_n, n > 50. Global optimization using the parameterized GAP model revealed energetically favorable Mg₅₁-Mg₅₃ clusters, showing early onset of pyramidal core formation previously reported only from Mg₅₄. Global optimization identified new global minima candidates for the "magic" Mg₅₉, Mg₆₉, Mg₇₄ and Mg₉₉ clusters. The presence of hcp-like motifs doesn’t significantly influence structural stability in clusters with Mg_n, n < 100, as no structural differences were observed between GM “magic” clusters and others of similar size.